This paper proposes a data-driven learning method to describe the personal thermal complaint behavior
in a complaint-driven environment control system. The complaint-driven system only uses personal
human complaints to control the personal office environment. It avoids the user’s direct control on the
set-point of the room, which usually results in unreasonable and uncomfortable set-point. A two-stage
classifier model is proposed, using personal thermal compliant data with respect to the transient and
steady complaint behaviors. The classifier structure is developed based on the properties of human
thermal perception with parameters to learn for different users. Quantitative results using experimental
data show that the model has lower false negative rate than traditional data-driven classification model
and acceptable false detection rate. Practical implementation and subjects’ questionnaire evaluation
demonstrate the satisfying performance of the model in real environment control. We also discuss the
limitations and potential extensions of the model at the end of this paper.